メニュー
Generates a self-contained HTML walkthrough of a codebase feature, flow, architecture, or schema. Triggers: "$walkthrough", "walk me through", "trace the code path", "explain this flow", "visualize the data model".
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
| name | walkthrough |
| description | Generates a self-contained HTML walkthrough of a codebase feature, flow, architecture, or schema. Triggers: "$walkthrough", "walk me through", "trace the code path", "explain this flow", "visualize the data model". |
| compatibility | Designed for Claude Code (or similar products). Requires a browser to open generated HTML files. |
| allowed-tools | Bash Read Write Glob Grep Task |
| metadata | {"author":"Alexander Opalic","version":"1.2"} |
Generate interactive HTML files that give new developers a quick mental model of how a feature or system works. The goal is fast onboarding — readable in under 2 minutes.
Three formats, each suited to a different kind of question:
| Format | Question it answers | Output |
|---|---|---|
| Architecture Flowchart | "What are the pieces and how are they wired?" | Mermaid flowchart, clickable nodes, detail panel |
| ER Diagram | "What's the data model?" | Mermaid ER diagram, entities + relationships |
| Narrated Trace | "Show me what happens, step by step, with real artifacts" | Linear timeline — each step shows the actual HTTP body / code / event / state / JSX |
CRITICAL: When this skill is triggered, you MUST generate a walkthrough-*.html file. Never respond with just text — always produce the interactive HTML output.
Always dark mode. Every walkthrough uses a pure black background (#000000), white text, and purple accents. Never generate light-mode walkthroughs. The <html> tag MUST include style="color-scheme: dark".
Choose by signal in the user's prompt. In the order checked:
$walkthrough (no topic).Ambiguous? If the request fits more than one (e.g., "explain how X works"), ask the user once — don't guess:
Which format would help most? A. Architecture flowchart — boxes + arrows showing the pieces and wiring B. Narrated trace — one real request walked through the code, step by step, with the actual artifact at each hop C. Both (architecture first, then trace) — generate two files
If you don't pick, I'll go with B for flow/lifecycle topics and A for system/layer topics.
After the user replies (or after applying the default), commit to one format and proceed. Do not switch mid-generation.
Frame the walkthrough as a mental model for someone new:
For traces, pick (or ask the user to pick) one realistic scenario — a specific user message, URL parameter, button click, or event payload. The whole trace stands or falls on this choice; pick the most representative path, not the simplest.
If the request is vague (and not a bare $walkthrough), ask one clarifying question. Otherwise proceed.
Always read real source files before generating. Never fabricate code paths.
Use the Task tool to delegate exploration to subagents. This keeps the main context clean for HTML generation and parallelizes the research phase.
Do a quick Glob/Grep yourself (1-2 calls max) to identify the relevant directories and file groups. Then split the exploration into 2-4 independent research tasks.
Use Task with subagent_type: "Explore" to launch multiple agents in a single message (parallel). Each agent investigates one area and reports back the purpose and connections of each piece — not code dumps.
Do NOT read the files yourself — let the subagents do it. Your job is to orchestrate and then synthesize their results into the HTML.
The shape of the report depends on the format:
Tell each subagent to return a structured report:
NODE: drawingInteraction
label: Drawing Interaction
file: app/features/tools/useDrawingInteraction.ts
purpose: Converts pointer events into shape data. The bridge between raw mouse input and the element model.
connects_to: canvasRenderer (feeds shape data), toolState (reads active tool)
key_snippet:
const element = createElement(tool, startPoint, currentPoint)
lang: typescript
Every node MUST include a code snippet. Pick the single most useful piece — a key function call, type definition, config line, or core algorithm (1–5 lines).
Tell each subagent to trace ONE concrete execution. Example prompt body:
Trace ONE concrete execution of
<scenario>through the codebase. Use a realistic input (e.g., a specific user message or URL param). For each step in execution order, report the artifact that exists at that moment:
- STEP
<N>kind: one ofHTTP REQUEST,AGENT WIRING,TYPESCRIPT PROGRAM,SANDBOX EXECUTION,SSE FRAME,STORE DISPATCH,STATE DIFF,REACT RENDER,LOG, etc.- file: path
- what happens in 1–2 sentences (plain English, why this hop exists)
- code: 4–12 lines showing the actual artifact at this point — real HTTP body, real function output, real JSON, real state mutation, real JSX. Real values, not
foo/bar/...placeholders.- lang: typescript / tsx / json / bash / text
Aim for 5–9 steps total. The reader should be able to read the code blocks alone, top-to-bottom, and watch the data transform.
If you split the trace across multiple subagents, give each a contiguous range of steps (e.g., "steps 1–3: client + server boot", "steps 4–6: sandbox + streaming", "steps 7–9: store + render") and tell them what the previous range produced as output.
Once subagents return, you have everything needed. Do NOT read more files or launch more subagents. Go directly to Step 4.
Combine results:
kind, title, description, files, lang, code).If a subagent's report is missing something, drop the affected node/step rather than reading files yourself — but ensure you still meet the format's minimum.
Create a single self-contained HTML file. Pick the right reference based on format:
| Format | Reference |
|---|---|
| Architecture Flowchart | references/html-patterns.md |
| ER Diagram | references/html-patterns.md |
| Narrated Trace | references/trace-patterns.md |
File location: Write to the project root as walkthrough-{topic}.html (e.g., walkthrough-canvas-drawing.html). Use kebab-case for the topic slug. For the "Both" choice, write two files: walkthrough-{topic}-architecture.html and walkthrough-{topic}-trace.html.
Architecture (all three formats):
<script type="module"> — native ES modules, no BabelReact.createElement(), no JSXRequired elements vary by format — see the reference file. Common to all:
vitesse-dark theme)lang field — never omit, default to "typescript"After writing the file, open it in the user's browser:
open walkthrough-{topic}.html # macOS
Use for: feature flows, request lifecycles, system overviews, "how the pieces connect".
Direction: graph TD (top-down) for hierarchical flows, graph LR (left-right) for sequential pipelines or request/response stories.
Node types (styled by category):
| Type | Style | Use for |
|---|---|---|
component | Purple-500 | UI components, pages |
composable | Purple-600 | Composables, hooks, bridges |
utility | Purple-700 | Utils, helpers, pure functions |
external | Gray-600 | Libraries, browser APIs, external services |
event | Purple-200 | Events, user actions, triggers |
data | Purple-600 | Stores, state, registries |
Subgraphs: Group related nodes into 2-4 subgraphs with approachable mental-model labels (e.g., "User Input", "Core Logic", "Visual Output") — not technical layer names. Numbered subgraphs (1 · User Input, 2 · Core Logic) help when the diagram is sequential.
Node IDs: Descriptive camelCase (e.g., drawingInteraction, canvasRenderer).
Node labels: Plain English ("Drawing Interaction"), not function signatures or file names.
Edges: Plain verbs ("triggers", "feeds into", "reads", "produces", "watches"). Not API method names.
| Mermaid syntax | Use for |
| -------------- | ------- | --- | ------------------------------------------------------------------ |
| A --> | "label" | B | Direct call / synchronous |
| A -.-> | "label" | B | Reactive / watch / subscribe |
| A ==> | "label" | B | Streamed event / emission (visually distinct, can be CSS-animated) |
Click binding (mandatory on every node):
click nodeId nodeClickHandler "View details"
where nodeClickHandler is a global JS function defined in the HTML.
Use for: schema design, table relationships, migrations, data models.
Syntax:
erDiagram
USERS {
string id PK
string email UK
string name
timestamp created_at
}
TEAMS {
string id PK
string name
string owner_id FK
}
USERS ||--o{ TEAMS : "owns"
Cardinality:
| Syntax | Meaning |
|---|---|
||--o{ | One to many |
||--|| | One to one |
}o--o{ | Many to many |
||--o| | One to zero-or-one |
Column markers: PK (primary), FK (foreign), UK (unique).
Click handlers: ER entities don't support Mermaid click callbacks — attach handlers manually via querySelectorAll('.entityLabel') after render. See html-patterns.md for the pattern.
Use for: "step by step", "what happens when X", request lifecycles, end-to-end flows. Output is a linear timeline — no diagram. See references/trace-patterns.md for the full template.
Step shape:
{
kind: 'SSE FRAME ON THE WIRE', // ≤ 4 words, uppercase
title: '5 · Each ui_* call emits an event',
description: '1-2 sentences in plain text. What and why.',
files: ['src/.../ui-bindings.ts'],
lang: 'json',
code: `data: { "type": "custom", "eventType": "storefront:ui", ... }`,
}
Suggested step kinds (mix and match — these are anchors, not a fixed taxonomy):
HTTP REQUEST · HTTP RESPONSE · AGENT WIRING · CONFIG · TYPESCRIPT PROGRAM · CODE PATH · SANDBOX EXECUTION · EXECUTION TRACE · LOG · SSE FRAME ON THE WIRE · EVENT · MESSAGE · STORE DISPATCH · STATE DIFF · REACT RENDER · DOM
Key constraints:
code is a real artifact, not a concept summary. If the step is "the model writes a program", the code IS the program — not a comment describing it.foo/bar/....N · Imperative phrase (numeric prefix is stripped at render time but kept in source for at-a-glance scanning).Before finishing, verify the format-appropriate items:
<html> tag includes style="color-scheme: dark"lang explicitly setclick nodeId nodeClickHandler "View details" binding||--o{, etc.).entityLabel click handlers attached after render (Mermaid click syntax doesn't work for ER)kind, title, description, files, lang, codecode block is a real artifact (HTTP body, code, log, JSON, JSX) — not a concept summaryfoo/bar)title is N · Imperative phrasekind pill is ≤ 4 words and uppercaseTurn a topic into a structured blog-post draft using Good Docs templates with blog voice. Triggers: "$good-docs-writer <intent> <topic>", "write a blog post about X", "draft a blog post on X".
Read/write brain files (Obsidian vault at brain/). Triggers: brain/ modifications, "add to brain".
Audit and evolve the brain vault — prune outdated content, discover cross-cutting principles, review skills for structural encoding opportunities. Triggers: "meditate", "audit the brain".
Break down medium-to-large tasks into phased plans in brain/plans/. Planning only — does not implement. Use for new features, multi-file refactors, or architectural changes — not small fixes. Triggers: "plan this", "break this down".
Reflect on the conversation and update the brain. Use when wrapping up, after mistakes or corrections, or when significant codebase knowledge was gained. Triggers: "reflect", "remember this".
Principle-grounded review of code changes, PRs, or plans. Use when asked to review, critique, or assess quality of work — "review", "review this", "code review", "check this".